Entry 2Bayesian Visual Analytics for Geo-Spatial Temporal Data
Bindu Gupta et al. — Tata Consultancy Services Research, India

Our approach utilises exploratory visualizations to understand the data and a vi- sual interface to a Bayesian Network model of the data for answering various types of questions. The main advantages of using Bayesian networks are that (a) dependence between variables is modelled by the network, (b) once the network has been learned it can be used for querying without touching the data, (c) probabilistic queries run faster on large amounts of data, (d) all queries can be answered probabilistically, even those would not yield results from data and (e) prediction of variables such as Cost Per Unit for new test centres can be done.

We provide exploratory visualizations of the data at state and region levels. A region is defined as the convex hull of all latitude, longitude pairs with the same first three digits in their zip codes and roughly map to cities. Line charts showing various attributes such as the national average cost per unit and utilization at different dates provide an overview. Users may select an attribute and move the time slider to view the corresponding heatmap for all states and regions, figure 1, A. Using small multiples attributes can be compared over time, figure 1, B & C.

We learn a Bayesian model on regional data which captures the conditional depen- dencies between the ‘cause variables’ such as Change in Number of Test Centers and Change in Capacity, and ‘effect variables’ such as Utilization, Cost per unit and Change in Demand. Using this model, we can add/delete test centers in regions and query to observe the effects in the same region and neighboring regions by comparing prior and posterior distribution of effect variables, figure 1, F. Further, we can query different combinations of ‘effect variables’ and visualise the changes in the regional distribution over the US map, figure 1, D &E.

Bindu Gupta, Gunjan Sehgal, Kaushal Paneri, Aditeya Pandey,
bindu.gupta2@tcs.com sehgal.gunjan@tcs.com kaushal.paneri@tcs.com
Karamjit Singh,Geetika Sharma,Gautam Shroff, 
Tata Consultancy Services Research, India